
# Prepare dataset
# Trait data (columns 2:9)
trait.data <- new.d

# write.csv(new.d,"Trait_data.csv")
# write.csv(mean_damage_df.1,"DamageS1.csv")
# write.csv(mean_damage_df.2,"DamageS2.csv")

# Damage data (columns 2:3)
damage_subset <- mean_damage_df.2

# Ensure response column is explicitly named
colnames(damage_subset)[1] <- "tip.label"  # Rename first column to 'response'
colnames(damage_subset)[3] <- "response"  # Rename first column to 'response'

# # Combine trait data and damage data into one dataset
# data <- cbind(damage_subset, trait.data)  # Ensure alignment of rows

# If there's an ID column, use merge instead:
data <- merge(damage_subset, trait.data, by = "tip.label")
data$tip.label <- as.factor(data$tip.label)

# Remove rows with NA values in either the response or predictors
data <- data %>% na.omit()

trait.data <- data[,c(12:19)]
# Generate all combinations of 1,2, and 3 traits
trait_combinations <- lapply(1:3, function(n) {
  combn(colnames(trait.data), n, simplify = FALSE)
}) %>%
  unlist(recursive = FALSE)  # Flatten the list of combinations

# Fit GLMs and store formulas and AIC values
results <- lapply(trait_combinations, function(predictor_set) {
  # Construct formula
  formula <- as.formula(paste("response ~", paste(predictor_set, collapse = " + ")))
  
  # Try fitting the model, handle errors gracefully
  tryCatch({
    model <- glm(formula, data = data,family = gaussian)  # Adjust family if necessary
    list(formula = formula, aic = AIC(model))  # Return formula and AIC
  }, error = function(e) {
    list(formula = formula, aic = NA)  # Return NA for failed models
  })
})

# Remove models with NA AICs
results <- results[!sapply(results, function(x) is.na(x$aic))]

# Extract AIC values and formulas into a data frame
results_df <- do.call(rbind, lapply(results, function(res) {
  data.frame(formula = deparse(res$formula), aic = res$aic)
}))

# Sort by AIC
results_df <- results_df %>% arrange(aic)

# Print the best model
cat("Best Model:\n")
print(results_df[1, ])

# Optional: Show all models sorted by AIC
cat("\nAll Models Sorted by AIC:\n")
print(results_df)

results_df <- results_df[-1,]
# Calculate ΔAIC
delta_aic <- results_df$aic - min(results_df$aic)
delta_aic

# Calculate Akaike weights
akaike_weights <- exp(-0.5 * delta_aic) / sum(exp(-0.5 * delta_aic))
akaike_weights

# Summary table
aic_table <- data.frame(
  Model = results_df$formula,
  AIC = results_df$aic,
  Delta_AIC = delta_aic,
  Akaike_Weight = akaike_weights
)

# Sort by AIC
aic_table <- aic_table[order(aic_table$AIC), ]
print(aic_table)

# Plot observed vs. predicted
# Best fit model:
model.3 <- glm(response ~ tlp + osmoticpot + lma, data = data,family=gaussian)
# McFadden's R²: Measures the improvement of the model 
# compared to a baseline (null) model. 
# It ranges from 0 to 1, where higher values indicate
# better fit.
# Calculate McFadden's R^2
1 - (model.3$deviance / model.3$null.deviance)
vif(model.3)

model.4 <- glm(response ~ P50 + hv + height, data = data,family=gaussian)
# McFadden's R²: Measures the improvement of the model 
# compared to a baseline (null) model. 
# It ranges from 0 to 1, where higher values indicate
# better fit.
# Calculate McFadden's R^2
1 - (model.4$deviance / model.4$null.deviance)
vif(model.4)
model.4$residuals
summary(model.4)
# 
# # Check for collinearity
# cor(data[, c("tlp","osmoticpot", "lma")])
# cor(data[, c("P50","hv", "height")])
# 
# # Check for NAs
# summary(data)
# 
# apply(data[, c("P50","hv", "height")], 2, sd)
# 
# # Check the residuals of the data
# plot(model$residuals)
# 
# # install.packages("car")
# vif(model)
# vif(model1)
# 
# plot(model, which = 1)  # Residuals vs. Fitted plot
# plot(model, which = 2) # Q - Q plot
# 
# 
# # Generate predicted values
# data$predicted <- predict(model, type = "response")
# 
# # Plot predicted vs observed
# ggplot(data, aes(x = predicted, y = response)) +
#   geom_point() +
#   geom_abline(intercept = 0, slope = 1, linetype = "dashed", color = "red") +
#   labs(x = "Predicted Values", y = "Observed Values", title = "Predicted vs. Observed Values") +
#   theme_minimal()
# 
# # Residuals vs fitted values plot
# data$residuals <- residuals(model, type = "pearson")
# 
# ggplot(data, aes(x = predicted, y = residuals)) +
#   geom_point() +
#   geom_hline(yintercept = 0, linetype = "dashed", color = "red") +
#   labs(x = "Fitted Values", y = "Residuals", title = "Residuals vs Fitted Values") +
#   theme_minimal()
# 
# Plot the effect of P50
p50_range <- seq(min(data$P50), max(data$P50), length.out = 100)
p50_data <- data.frame(P50 = p50_range, hv = mean(data$hv), height = mean(data$height))
p50_data$response <- predict(model.4, newdata = p50_data, type = "response")

ggplot(p50_data, aes(x = P50, y = response)) +
  geom_line(color = "blue") +
  labs(x = "P50", y = "Predicted Response", title = "Effect of P50 on Response") +
  theme_minimal()

# Plot the effect of hv
hv_range <- seq(min(data$hv), max(data$hv), length.out = 100)
hv_data <- data.frame(hv = hv_range, P50 = mean(data$P50), height = mean(data$height))
hv_data$response <- predict(model.4, newdata = hv_data, type = "response")

ggplot(hv_data, aes(x = hv, y = response)) +
  geom_line(color = "red") +
  labs(x = "hv", y = "Predicted Response", title = "Effect of hv on Response") +
  theme_minimal()

# Plot the effect of height
h_range <- seq(min(data$height), max(data$height), length.out = 100)
h_data <- data.frame(height = h_range, P50 = mean(data$P50), hv = mean(data$hv))
h_data$response <- predict(model.4, newdata = h_data, type = "response")

ggplot(h_data, aes(x = height, y = response)) +
  geom_line(color = "purple") +
  labs(x = "height", y = "Predicted Response", title = "Effect of moe on Response") +
  theme_minimal()